<![CDATA[A stroke is a sudden onset of brain dysfunction, lasting for 24 hours or longer, resulting from clinically focal and global brain dysfunction. As many as 15 million people die from stroke each year. The stroke patients need an immediate treatment to minimize the risk of brain damage. One of the proponents for the stroke diagnosis is through a computed tomography (CT) image. In recent years, the image processing techniques capable to detect stroke patterns in a brain image, it can be useful for doctors and radiologists in doing diagnosis and treatment. This study aims to compare the level of accuracy using augmentation and without augmentation and hyperparameters using the Convolutional Neural Network in the EfficientNet-B0 architecture to classify ischemic, hemorrhagic, and normal brain stroke images. The data augmentation is produced by rotating, horizontal flipping, and contrast tuning of the original data. Testing data is provided as much as 20% of the portion of the original and augmented data, and the other 80% is used for the training process to find the optimal model. The model search is based on the composition of the training and validation data with a ratio of 70:30, 80:20 and 90:10. The experimental results show that the best performance is obtained for the combined original and augmented images, with accuracies of 97%, 93%, and 94%, respectively, for the three types of data-test: original, augmented, and combined. The merging of original and augmentated images for training data has shown that the model is robust enough in producing high accuracy results.]]>
Copyrights © 2023
